Abstract: Based on the spectral and shape features of loess landslide scarp, the authors segmented the high resolution remote sensing image by using regional growing and merging method, and extracted the landslide scarp area from the segmented image. The skeleton of the scarp area was obtained by way of the binary composition to characterize loess landslide. Firstly, The typical semi-circle or round-backed armchair shape characteristics and spectral characteristics became one of the important marks for loess landslide remote sensing interpretation, whereas the initial shape of the scarp was achieved by taking advantage of the regional growing and merging method with the growing parameter being 47.5 and the merging parameter being 49. Then, the holes of initial scarp of the scarp generated by segmentation process were filled. Finally, the main framework of landslide scarp was extracted using binarization skeleton algorithm, which constituted the mark of the loess landslide. This method is an improved and enhanced means for loess landslide expression in the remote sensing interpretation.